Colab 未使用 GPU 进行培训

Question

I'm training a custom model on GTSRB dataset for traffic sign recognition on Google Colab. I've successfully built my model, But when I'm training my model it runs on CPU instead of GPU. I've selected GPU runtime earlier and using mirrored strategy using keras-Tensorflow. Any help will be appreciated. I'm adding snippets of my code.

with strategy.scope():

  test_datagen = ImageDataGenerator(rescale=1./255)
  train_datagen = ImageDataGenerator(rescale=1./255, shear_range=0.2, zoom_range=0.2, 
  horizontal_flip=False)
  train_generator = train_datagen.flow_from_directory('/content/drive/MyDrive/Train',target_size= 
  (64, 64),batch_size=32,class_mode='categorical')
  validation_generator = 
  test_datagen.flow_from_directory(directory='/content/drive/MyDrive/validation', target_size= 
  (64, 64), batch_size=32, class_mode='categorical')
  checkpoint_path = "training_1/cp.ckpt"
  cp_callback = ModelCheckpoint(filepath=checkpoint_path, save_weights_only=True, verbose=1, 
  save_freq='epoch')

with strategy.scope():
  model = training(classes)
  model.compile(loss='categorical_crossentropy', optimizer=Adam(lr=1e-4),metrics=['accuracy'])
model.fit(train_generator, steps_per_epoch=200, epochs=50, validation_data=validation_generator, 
validation_steps=100, callbacks=[cp_callback])

while training I've checked the GPU and CPU usage which is shown in the image:

Answer 1

Check the status of the GPU while the model is being run using the command nvidia-smi .

Answer 2

I have had this before multiple times with multiple platforms (Kaggle, Colab) but only managed to solve this once. In that case, it was due to tensorflow_gpu (gpu version of tensorflow) not being installed when I switched tensorflow versions. You can check the version of tensorflow_gpu and the version of tensorflow that is installed. Make sure that the versions are the same. This may be totally off the mark or it might work for you.